Pegi1727/GenAI-Identity-Compression-Framework: `Dataset for the Study of Generative AI's Impact on LL2 Writing Identity and Agency: The Authenticity Gap Framework`

Abstract

Description: This dataset accompanies the research article titled "When AI Improves the Text but Changes the Voice: Generative AI, Authorial Identity, and Authenticity in EFL Writing." The repository contains a comprehensive set of anonymized data and analysis materials investigating the tension between AI-mediated linguistic optimization and authorial voice. Contents include: Anonymized Corpus: Paired L2 essays (original independent versions vs. GPT-4o revised versions) from 60 advanced EFL learners. Quantitative Metrics: Linguistic features including MTLD (Measure of Textual Lexical Diversity), T-unit analysis, and stance marker frequencies. Psychological Scales: Raw and processed scores from the Authenticity Gap Scale (AGS). Analysis Scripts: R and Python code used for Linear Mixed-Effects Models (LME) and stylistic homogenization analysis. Research Instruments: The standardized prompt used for AI revision and the semi-structured interview protocol. Keywords: Generative AI, L2 Writing, Authorial Identity, Authenticity Gap, Linguistic Agency, EFL.